OK - based on your new flight description and thus discarding the GPS data entirely, we can estimate position purely from the two IMU datasets:
View attachment 54807
They don't agree perfectly, probably due to bias uncertainties, but they broadly agree on a flight path consistent with the one that you described. That's also consistent with scrambled GPS reception, most likely due to building reflections and poor sky view:
View attachment 54808
As a result, it appears that the FC was completely confused by the position data it was receiving, and was heading north when it thought it was heading south. That's a hazard of launching with poor GPS reception in a location with a very poor sky view and multiple vertical reflectors (buildings).